platform/kernel/linux-rpi.git
2 years agoKVM: selftests: Convert evmcs_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:57:49 +0000 (16:57 -0800)]
KVM: selftests: Convert evmcs_test away from VCPU_ID

Convert evmcs_test to use vm_create_with_one_vcpu() and pass around a
'struct kvm_vcpu' object instead of using a global VCPU_ID.  Note, this is
a "functional" change in the sense that the test now creates a vCPU with
vcpu_id==0 instead of vcpu_id==5.  The non-zero VCPU_ID was 100% arbitrary
and added little to no validation coverage.  If testing non-zero vCPU IDs
is desirable for generic tests, that can be done in the future by tweaking
the VM creation helpers.

Opportunistically use vcpu_run() instead of _vcpu_run(), the test expects
KVM_RUN to succeed.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert hyperv_clock away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:54:04 +0000 (16:54 -0800)]
KVM: selftests: Convert hyperv_clock away from VCPU_ID

Convert hyperv_clock to use vm_create_with_one_vcpu() and pass around a
'struct kvm_vcpu' object instead of using a global VCPU_ID.

Opportunistically use vcpu_run() instead of _vcpu_run() with an open
coded assert that KVM_RUN succeeded.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert hyperv_features away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:50:11 +0000 (16:50 -0800)]
KVM: selftests: Convert hyperv_features away from VCPU_ID

Convert hyperv_features to use vm_create_with_one_vcpu() and pass around
a 'struct kvm_vcpu' object instead of using a global VCPU_ID.

Opportunistically use vcpu_run() instead of _vcpu_run() with an open
coded assert that KVM_RUN succeeded.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert hyperv_svm_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:47:48 +0000 (16:47 -0800)]
KVM: selftests: Convert hyperv_svm_test away from VCPU_ID

Convert hyperv_svm_test to use vm_create_with_one_vcpu() and pass around a
'struct kvm_vcpu' object instead of using a global VCPU_ID.  Note, this is
a "functional" change in the sense that the test now creates a vCPU with
vcpu_id==0 instead of vcpu_id==1.  The non-zero VCPU_ID was 100% arbitrary
and added little to no validation coverage.  If testing non-zero vCPU IDs
is desirable for generic tests, that can be done in the future by tweaking
the VM creation helpers.

Opportunistically use vcpu_run() instead of _vcpu_run(), the test expects
KVM_RUN to succeed.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert kvm_clock_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:46:56 +0000 (16:46 -0800)]
KVM: selftests: Convert kvm_clock_test away from VCPU_ID

Convert kvm_clock_test to use vm_create_with_one_vcpu() and pass around a
'struct kvm_vcpu' object instead of using a global VCPU_ID.

Opportunistically use vcpu_run() instead of _vcpu_run() with an open
coded assert that KVM_RUN succeeded.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert tsc_msrs_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:45:15 +0000 (16:45 -0800)]
KVM: selftests: Convert tsc_msrs_test away from VCPU_ID

Convert tsc_msrs_test to use vm_create_with_one_vcpu() and pass around a
'struct kvm_vcpu' object instead of using a global VCPU_ID.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert vmx_exception_with_invalid_guest_state away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:44:36 +0000 (16:44 -0800)]
KVM: selftests: Convert vmx_exception_with_invalid_guest_state away from VCPU_ID

Convert vmx_exception_with_invalid_guest_state to use
vm_create_with_one_vcpu() and pass around a 'struct kvm_vcpu' object
instead of using a global VCPU_ID.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert userspace_msr_exit_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:38:01 +0000 (16:38 -0800)]
KVM: selftests: Convert userspace_msr_exit_test away from VCPU_ID

Convert userspace_msr_exit_test to use vm_create_with_one_vcpu() and pass
around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.
Note, this is a "functional" change in the sense that the test now
creates a vCPU with vcpu_id==0 instead of vcpu_id==1.  The non-zero
VCPU_ID was 100% arbitrary and added little to no validation coverage.
If testing non-zero vCPU IDs is desirable for generic tests, that can be
done in the future by tweaking the VM creation helpers.

Opportunistically use vcpu_run() instead of _vcpu_run() with an open
coded assert that KVM_RUN succeeded.  Fix minor coding style violations
too.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert vmx_apic_access_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:27:47 +0000 (16:27 -0800)]
KVM: selftests: Convert vmx_apic_access_test away from VCPU_ID

Convert vmx_apic_access_test to use vm_create_with_one_vcpu() and pass
around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.

Opportunistically make the "vm" variable local, it is unused outside of
main().

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert vmx_close_while_nested_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:26:58 +0000 (16:26 -0800)]
KVM: selftests: Convert vmx_close_while_nested_test away from VCPU_ID

Convert vmx_close_while_nested_test to use vm_create_with_one_vcpu() and
pass around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.
Note, this is a "functional" change in the sense that the test now
creates a vCPU with vcpu_id==0 instead of vcpu_id==5.  The non-zero
VCPU_ID was 100% arbitrary and added little to no validation coverage.
If testing non-zero vCPU IDs is desirable for generic tests, that can be
done in the future by tweaking the VM creation helpers.

Opportunistically make the "vm" variable local, it is unused outside of
main().

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert vmx_dirty_log_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:26:03 +0000 (16:26 -0800)]
KVM: selftests: Convert vmx_dirty_log_test away from VCPU_ID

Convert vmx_dirty_log_test to use vm_create_with_one_vcpu() and pass
around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.
Note, this is a "functional" change in the sense that the test now
creates a vCPU with vcpu_id==0 instead of vcpu_id==1.  The non-zero
VCPU_ID was 100% arbitrary and added little to no validation coverage.
If testing non-zero vCPU IDs is desirable for generic tests, that can be
done in the future by tweaking the VM creation helpers.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert set_sregs_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:24:50 +0000 (16:24 -0800)]
KVM: selftests: Convert set_sregs_test away from VCPU_ID

Convert set_sregs_test to use vm_create_with_one_vcpu() and pass around
a 'struct kvm_vcpu' object instead of using a global VCPU_ID.  Note, this
is a "functional" change in the sense that the test now creates a vCPU
with vcpu_id==0 instead of vcpu_id==5.  The non-zero VCPU_ID was 100%
arbitrary and added little to no validation coverage.  If testing
non-zero vCPU IDs is desirable for generic tests, that can be
done in the future by tweaking the VM creation helpers.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert vmx_nested_tsc_scaling_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:23:43 +0000 (16:23 -0800)]
KVM: selftests: Convert vmx_nested_tsc_scaling_test away from VCPU_ID

Convert vmx_nested_tsc_scaling_test to use vm_create_with_one_vcpu() and
pass around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert platform_info_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:11:59 +0000 (16:11 -0800)]
KVM: selftests: Convert platform_info_test away from VCPU_ID

Convert platform_info_test to use vm_create_with_one_vcpu() and pass
around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert kvm_pv_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:10:04 +0000 (16:10 -0800)]
KVM: selftests: Convert kvm_pv_test away from VCPU_ID

Convert kvm_pv_test to use vm_create_with_one_vcpu() and pass arounda
'struct kvm_vcpu' object instead of using a global VCPU_ID.

Opportunistically use vcpu_run() instead of _vcpu_run() with an open
coded assert that KVM_RUN succeeded.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert hyperv_cpuid away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:08:48 +0000 (16:08 -0800)]
KVM: selftests: Convert hyperv_cpuid away from VCPU_ID

Convert hyperv_cpuid to use vm_create_with_one_vcpu() and pass around a
'struct kvm_vcpu' object instead of using a global VCPU_ID.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert sync_regs_test away from VCPU_ID
Sean Christopherson [Wed, 16 Feb 2022 00:04:25 +0000 (16:04 -0800)]
KVM: selftests: Convert sync_regs_test away from VCPU_ID

Convert sync_regs_test to use vm_create_with_one_vcpu() and pass around
a 'struct kvm_vcpu' object instead of using a global VCPU_ID.  Note, this
is a "functional" change in the sense that the test now creates a vCPU
with vcpu_id==0 instead of vcpu_id==5.  The non-zero VCPU_ID was 100%
arbitrary and added little to no validation coverage.  If testing
non-zero vCPU IDs is desirable for generic tests, that can be
done in the future by tweaking the VM creation helpers.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert svm_vmcall_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 23:56:54 +0000 (15:56 -0800)]
KVM: selftests: Convert svm_vmcall_test away from VCPU_ID

Convert svm_vmcall_test to use vm_create_with_one_vcpu() and pass around
a 'struct kvm_vcpu' object instead of using a global VCPU_ID.  Note, this
is a "functional" change in the sense that the test now creates a vCPU
with vcpu_id==0 instead of vcpu_id==5.  The non-zero VCPU_ID was 100%
arbitrary and added little to no validation coverage.  If testing
non-zero vCPU IDs is desirable for generic tests, that can be done in the
future by tweaking the VM creation helpers.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert svm_int_ctl_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 23:55:26 +0000 (15:55 -0800)]
KVM: selftests: Convert svm_int_ctl_test away from VCPU_ID

Convert svm_int_ctl_test to use vm_create_with_one_vcpu() and pass around
a 'struct kvm_vcpu' object instead of using a global VCPU_ID.

Opportunistically make the "vm" variable a local function variable, there
are no users outside of main().

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert state_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 23:53:16 +0000 (15:53 -0800)]
KVM: selftests: Convert state_test away from VCPU_ID

Convert state_test to use vm_create_with_one_vcpu() and
vm_recreate_with_one_vcpu(), and pass around a 'struct kvm_vcpu' object
instead of using a global VCPU_ID.  Note, this is a "functional" change
in the sense that the test now creates a vCPU with vcpu_id==0 instead of
vcpu_id==5.  The non-zero VCPU_ID was 100% arbitrary and added little to
no validation coverage.  If testing non-zero vCPU IDs is desirable for
generic tests, that can be done in the future by tweaking the VM creation
helpers.

Opportunistically use vcpu_run() instead of _vcpu_run(), the test expects
KVM_RUN to succeed.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert smm_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 23:51:19 +0000 (15:51 -0800)]
KVM: selftests: Convert smm_test away from VCPU_ID

Convert smm_test to use vm_create_with_one_vcpu() and pass around a
'struct kvm_vcpu' object instead of using a global VCPU_ID.  Note, this
is a "functional" change in the sense that the test now creates a vCPU
with vcpu_id==0 instead of vcpu_id==1.  The non-zero VCPU_ID was 100%
arbitrary and added little to no validation coverage.  If testing
non-zero vCPU IDs is desirable for generic tests, that can be done in the
future by tweaking the VM creation helpers.

Opportunistically use vcpu_run() instead of _vcpu_run(), the test expects
KVM_RUN to succeed.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert pmu_event_filter_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 23:39:45 +0000 (15:39 -0800)]
KVM: selftests: Convert pmu_event_filter_test away from VCPU_ID

Convert pmu_event_filter_test to use vm_create_with_one_vcpu() and pass
around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.
Rename run_vm_to_sync() to run_vcpu_to_sync() accordingly.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert mmu_role_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 23:28:49 +0000 (15:28 -0800)]
KVM: selftests: Convert mmu_role_test away from VCPU_ID

Convert mmu_role_test to use vm_create_with_one_vcpu() and pass around
a 'struct kvm_vcpu' object instead of using a global VCPU_ID.
Note, this is a "functional" change in the sense that the test now
creates a vCPU with vcpu_id==0 instead of vcpu_id==1.  The non-zero
VCPU_ID was 100% arbitrary and added little to no validation coverage.
If testing non-zero vCPU IDs is desirable for generic tests, that can be
done in the future by tweaking the VM creation helpers.

Opportunistically use vcpu_run() instead of _vcpu_run() plus an open
coded assert that KVM_RUN succeeded.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert vmx_tsc_adjust_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 23:22:03 +0000 (15:22 -0800)]
KVM: selftests: Convert vmx_tsc_adjust_test away from VCPU_ID

Convert vmx_tsc_adjust_test to use vm_create_with_one_vcpu() and pass
around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.
Note, this is a "functional" change in the sense that the test now
creates a vCPU with vcpu_id==0 instead of vcpu_id==5.  The non-zero
VCPU_ID was 100% arbitrary and added little to no validation coverage.
If testing non-zero vCPU IDs is desirable for generic tests, that can be
done in the future by tweaking the VM creation helpers.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert vmx_set_nested_state_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 21:56:58 +0000 (13:56 -0800)]
KVM: selftests: Convert vmx_set_nested_state_test away from VCPU_ID

Convert vmx_set_nested_state_test to use vm_create_with_one_vcpu() and
pass around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.
Note, this is a "functional" change in the sense that the test now
creates a vCPU with vcpu_id==0 instead of vcpu_id==5.  The non-zero
VCPU_ID was 100% arbitrary and added little to no validation coverage.
If testing non-zero vCPU IDs is desirable for generic tests, that can be
done in the future by tweaking the VM creation helpers.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert vmx_pmu_msrs_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 21:53:06 +0000 (13:53 -0800)]
KVM: selftests: Convert vmx_pmu_msrs_test away from VCPU_ID

Convert vmx_pmu_msrs_test to use vm_create_with_one_vcpu() and pass
around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert vmx_preemption_timer_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 21:51:14 +0000 (13:51 -0800)]
KVM: selftests: Convert vmx_preemption_timer_test away from VCPU_ID

Convert vmx_preemption_timer_test to use vm_create_with_one_vcpu() and
pass around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.
Note, this is a "functional" change in the sense that the test now
creates a vCPU with vcpu_id==0 instead of vcpu_id==5.  The non-zero
VCPU_ID was 100% arbitrary and added little to no validation coverage.
If testing non-zero vCPU IDs is desirable for generic tests, that can be
done in the future by tweaking the VM creation helpers.

Opportunistically use vcpu_run() instead of _vcpu_run(), the test expects
KVM_RUN to succeed.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert xss_msr_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 21:49:19 +0000 (13:49 -0800)]
KVM: selftests: Convert xss_msr_test away from VCPU_ID

Convert xss_msr_test to use vm_create_with_one_vcpu() and pass around a
'struct kvm_vcpu' object instead of using a global VCPU_ID.  Note, this
is a "functional" change in the sense that the test now creates a vCPU
with vcpu_id==0 instead of vcpu_id==1.  The non-zero VCPU_ID was 100%
arbitrary and added little to no validation coverage.  If testing
non-zero vCPU IDs is desirable for generic tests, that can be done in the
future by tweaking the VM creation helpers.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert rseq_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 21:36:46 +0000 (13:36 -0800)]
KVM: selftests: Convert rseq_test away from VCPU_ID

Convert rseq_test to use vm_create_with_one_vcpu() and pass around a
'struct kvm_vcpu' object instead of using a global VCPU_ID.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Convert memslot_perf_test away from VCPU_ID
Sean Christopherson [Tue, 15 Feb 2022 21:34:31 +0000 (13:34 -0800)]
KVM: selftests: Convert memslot_perf_test away from VCPU_ID

Convert memslot_perf_test to use __vm_create_with_one_vcpu() and pass
around a 'struct kvm_vcpu' object instead of using a global VCPU_ID.
This is the first of many, many steps towards eliminating VCPU_ID from
all KVM selftests, and towards eventually purging the VM+vcpu_id mess.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Return the created vCPU from vm_vcpu_add()
Sean Christopherson [Wed, 16 Feb 2022 00:16:45 +0000 (16:16 -0800)]
KVM: selftests: Return the created vCPU from vm_vcpu_add()

Return the created vCPU from vm_vcpu_add() so that callers don't need to
manually retrieve the vCPU that was just added.  Opportunistically drop
the "heavy" function comment, it adds a lot of lines of "code" but not
much value, e.g. it's pretty obvious that @vm is a virtual machine...

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Rename 'struct vcpu' to 'struct kvm_vcpu'
Sean Christopherson [Tue, 15 Feb 2022 23:24:21 +0000 (15:24 -0800)]
KVM: selftests: Rename 'struct vcpu' to 'struct kvm_vcpu'

Rename 'struct vcpu' to 'struct kvm_vcpu' to align with 'struct kvm_vm'
in the selftest, and to give readers a hint that the struct is specific
to KVM.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Rename vcpu.state => vcpu.run
Sean Christopherson [Tue, 15 Feb 2022 21:26:01 +0000 (13:26 -0800)]
KVM: selftests: Rename vcpu.state => vcpu.run

Rename the "state" field of 'struct vcpu' to "run".  KVM calls it "run",
the struct name is "kvm_run", etc...

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Rename xAPIC state test's vcpu struct
Sean Christopherson [Mon, 18 Apr 2022 17:08:40 +0000 (10:08 -0700)]
KVM: selftests: Rename xAPIC state test's vcpu struct

Rename xapic_state_test's kvm_vcpu struct to xapic_vcpu to avoid a
collision when the common 'struct vcpu' is renamed to 'struct kvm_vcpu'
in a future patch.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Make vm_create() a wrapper that specifies VM_MODE_DEFAULT
Sean Christopherson [Tue, 19 Apr 2022 21:21:38 +0000 (14:21 -0700)]
KVM: selftests: Make vm_create() a wrapper that specifies VM_MODE_DEFAULT

Add ____vm_create() to be the innermost helper, and turn vm_create() into
a wrapper the specifies VM_MODE_DEFAULT.  Most of the vm_create() callers
just want the default mode, or more accurately, don't care about the mode.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Rename vm_create_without_vcpus() => vm_create()
Sean Christopherson [Tue, 19 Apr 2022 21:16:34 +0000 (14:16 -0700)]
KVM: selftests: Rename vm_create_without_vcpus() => vm_create()

Rename vm_create_without_vcpus() to vm_create() so that it's not
misconstrued as helper that creates a VM that can never have vCPUs, as
opposed to a helper that "just" creates a VM without vCPUs added at time
zero.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Rename vm_create() => vm_create_barebones(), drop param
Sean Christopherson [Mon, 18 Apr 2022 23:26:06 +0000 (16:26 -0700)]
KVM: selftests: Rename vm_create() => vm_create_barebones(), drop param

Rename vm_create() to vm_create_barebones() and drop the @phys_pages
param.  Pass '0' for the number of pages even though some callers pass
'DEFAULT_GUEST_PHY_PAGES', as the intent behind creating truly barebones
VMs is purely to create a VM, i.e. there aren't vCPUs, there's no guest
code loaded, etc..., and so there is nothing that will ever need or
consume guest memory.

Freeing up the name vm_create() will allow using the name for an inner
helper to the other VM creators, which need a "full" VM.

Opportunisticaly rewrite the function comment for addr_gpa2alias() to
focus on what the _function_ does, not what its _sole caller_ does.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Avoid memory allocations when adding vCPU in get-reg-list
Sean Christopherson [Thu, 9 Jun 2022 17:19:01 +0000 (10:19 -0700)]
KVM: selftests: Avoid memory allocations when adding vCPU in get-reg-list

Open code adding and doing setup for a vCPU in get-reg-list in order to
avoid the stack allocation that comes with aarch64_vcpu_add_default().
get-reg-list doesn't need to run the vCPU, and so doesn't need the guest
to be backed with memory.  This will allow future cleanup to turn what is
current vm_create() into a barebones helper.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Use vm_create_without_vcpus() in psci_test
Sean Christopherson [Mon, 18 Apr 2022 20:16:39 +0000 (13:16 -0700)]
KVM: selftests: Use vm_create_without_vcpus() in psci_test

Use vm_create_without_vcpus() instead of open coding a rough equivalent
in psci_test.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Use vm_create_without_vcpus() in hardware_disable_test
Sean Christopherson [Mon, 18 Apr 2022 20:16:32 +0000 (13:16 -0700)]
KVM: selftests: Use vm_create_without_vcpus() in hardware_disable_test

Use vm_create_without_vcpus() instead of open coding a rough equivalent
in hardware_disable_test.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Use vm_create_without_vcpus() in dirty_log_test
Sean Christopherson [Mon, 18 Apr 2022 20:02:55 +0000 (13:02 -0700)]
KVM: selftests: Use vm_create_without_vcpus() in dirty_log_test

Use vm_create_without_vcpus() instead of open coding a rough equivalent.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Use vm_create_without_vcpus() in set_boot_cpu_id
Sean Christopherson [Mon, 18 Apr 2022 20:00:58 +0000 (13:00 -0700)]
KVM: selftests: Use vm_create_without_vcpus() in set_boot_cpu_id

Use vm_create_without_vcpus() in set_boot_cpu_id instead of open coding
the equivlant now that the "without_vcpus" variant does
vm_adjust_num_guest_pages().

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Push vm_adjust_num_guest_pages() into "w/o vCPUs" helper
Sean Christopherson [Mon, 18 Apr 2022 19:58:44 +0000 (12:58 -0700)]
KVM: selftests: Push vm_adjust_num_guest_pages() into "w/o vCPUs" helper

Move the call to vm_adjust_num_guest_pages() from vm_create_with_vcpus()
down into vm_create_without_vcpus().  This will allow a future patch to
make the "w/o vCPUs" variant the common inner helper, e.g. so that the
"with_vcpus" helper calls the "without_vcpus" helper, instead of having
them be separate paths.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Add vm_create_*() variants to expose/return 'struct vcpu'
Sean Christopherson [Tue, 15 Feb 2022 23:14:51 +0000 (15:14 -0800)]
KVM: selftests: Add vm_create_*() variants to expose/return 'struct vcpu'

Add VM creation helpers to expose/return 'struct vcpu' so that tests
don't have to hardcode a VCPU_ID or make assumptions about what vCPU ID
is used by the framework just to retrieve a vCPU the test created.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftest: Add proper helpers for x86-specific save/restore ioctls
Sean Christopherson [Thu, 2 Jun 2022 17:30:06 +0000 (10:30 -0700)]
KVM: selftest: Add proper helpers for x86-specific save/restore ioctls

Add helpers for the various one-off helpers used by x86's vCPU state
save/restore helpers, and convert the other open coded ioctl()s to use
existing helpers.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Rename MP_STATE and GUEST_DEBUG helpers for consistency
Sean Christopherson [Thu, 2 Jun 2022 17:25:56 +0000 (10:25 -0700)]
KVM: selftests: Rename MP_STATE and GUEST_DEBUG helpers for consistency

Move the get/set part of the MP_STATE and GUEST_DEBUG helpers to the end
to align with the many other ioctl() wrappers/helpers.  Note, this is not
an endorsement of the predominant style, the goal is purely to provide
consistency in the selftests.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Dedup MSR index list helpers, simplify dedicated test
Sean Christopherson [Thu, 2 Jun 2022 19:24:18 +0000 (12:24 -0700)]
KVM: selftests: Dedup MSR index list helpers, simplify dedicated test

Consolidate the helper for retrieving the list of save/restore MSRs and
the list of feature MSRs, and use the common helpers in the related
get_msr_index_features test.  Switching to the common helpers eliminates
the testcase that KVM returns the same -E2BIG result if the input number
of MSRs is '1' versus '0', but considered that testcase isn't very
interesting, e.g. '0' and '1' are equally arbitrary, and certainly not
worth the additional code.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Harden and comment XSS / KVM_SET_MSRS interaction
Sean Christopherson [Thu, 2 Jun 2022 21:12:22 +0000 (14:12 -0700)]
KVM: selftests: Harden and comment XSS / KVM_SET_MSRS interaction

Assert that KVM_SET_MSRS returns '0' or '1' when setting XSS to a
non-zero value.  The ioctl() itself should "succeed", its only the
setting of the XSS MSR that should fail/fault.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Cache list of MSRs to save/restore
Sean Christopherson [Thu, 2 Jun 2022 21:12:14 +0000 (14:12 -0700)]
KVM: selftests: Cache list of MSRs to save/restore

Cache the list of MSRs to save restore, mostly to justify not freeing the
list in the caller, which simplifies consumption of the list.

Opportunistically move the XSS test's so called is_supported_msr() to
common code as kvm_msr_is_in_save_restore_list().  The XSS is "supported"
by KVM, it's simply not in the save/restore list because KVM doesn't yet
allow a non-zero value.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Simplify KVM_ENABLE_CAP helper APIs
Sean Christopherson [Thu, 2 Jun 2022 20:19:09 +0000 (13:19 -0700)]
KVM: selftests: Simplify KVM_ENABLE_CAP helper APIs

Rework the KVM_ENABLE_CAP helpers to take the cap and arg0; literally
every current user, and likely every future user, wants to set 0 or 1
arguments and nothing else.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Consolidate KVM_ENABLE_CAP usage
Sean Christopherson [Thu, 2 Jun 2022 19:55:59 +0000 (12:55 -0700)]
KVM: selftests: Consolidate KVM_ENABLE_CAP usage

Add __vm_enable_cap() and use it for negative tests that expect
KVM_ENABLE_CAP to fail.  Opportunistically clean up the MAX_VCPU_ID test
error messages.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Add a VM backpointer to 'struct vcpu'
Sean Christopherson [Tue, 15 Feb 2022 23:14:29 +0000 (15:14 -0800)]
KVM: selftests: Add a VM backpointer to 'struct vcpu'

Add a backpointer to 'struct vcpu' so that tests can get at the owning
VM when passing around a vCPU object.  Long term, this will be little
more than a nice-to-have feature, but in the short term it is a critical
step toward purging the VM+vcpu_id ioctl mess without introducing even
more churn.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Dedup vgic_init's asserts and improve error messages
Sean Christopherson [Thu, 9 Jun 2022 20:20:53 +0000 (13:20 -0700)]
KVM: selftests: Dedup vgic_init's asserts and improve error messages

Move the asserts for the many REDIST_REGS accesses into common helpers
instead of copy+pasting the same, unhelpful asserts over and over.  Not
providing the actual (or expected) value makes it unnecessarily painful
to debug failures, especially since test_assert() prints the errno
unconditionally, e.g. on success, it may print a stale, misleading errno.

Use kvm_device_attr_get() to handle the "success" check so that the
"success" and "expected == actual" asserts are separated, which will make
it far less likely that a user incorrectly assumes the ioctl() failed.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Split get/set device_attr helpers
Sean Christopherson [Thu, 9 Jun 2022 20:06:57 +0000 (13:06 -0700)]
KVM: selftests: Split get/set device_attr helpers

Split the get/set device_attr helpers instead of using a boolean param to
select between get and set.  Duplicating upper level wrappers is a very,
very small price to pay for improved readability, and having constant (at
compile time) inputs will allow the selftests framework to sanity check
ioctl() invocations.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Drop 'int' return from asserting *_has_device_attr()
Sean Christopherson [Fri, 18 Feb 2022 00:33:21 +0000 (16:33 -0800)]
KVM: selftests: Drop 'int' return from asserting *_has_device_attr()

Drop 'int' returns from *_device_has_attr() helpers that assert the
return is '0', there's no point in returning '0' and "requiring" the
caller to perform a redundant assertion.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Rename KVM_HAS_DEVICE_ATTR helpers for consistency
Sean Christopherson [Thu, 3 Mar 2022 03:49:16 +0000 (19:49 -0800)]
KVM: selftests: Rename KVM_HAS_DEVICE_ATTR helpers for consistency

Rename kvm_device_check_attr() and its variants to kvm_has_device_attr()
to be consistent with the ioctl names and with other helpers in the KVM
selftests framework.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Multiplex return code and fd in __kvm_create_device()
Sean Christopherson [Thu, 17 Feb 2022 20:21:33 +0000 (12:21 -0800)]
KVM: selftests: Multiplex return code and fd in __kvm_create_device()

Multiplex the return value and fd (on success) in __kvm_create_device()
to mimic common library helpers that return file descriptors, e.g. open().

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Move KVM_CREATE_DEVICE_TEST code to separate helper
Sean Christopherson [Thu, 17 Feb 2022 20:16:20 +0000 (12:16 -0800)]
KVM: selftests: Move KVM_CREATE_DEVICE_TEST code to separate helper

Move KVM_CREATE_DEVICE_TEST to its own helper, identifying "real" versus
"test" device creation based on a hardcoded boolean buried in the middle
of a param list is painful for readers.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Drop @test param from kvm_create_device()
Sean Christopherson [Thu, 17 Feb 2022 20:09:28 +0000 (12:09 -0800)]
KVM: selftests: Drop @test param from kvm_create_device()

Remove the two calls that pass @test=true to kvm_create_device() and drop
the @test param entirely.  The two removed calls don't check the return
value of kvm_create_device(), so other than verifying KVM doesn't explode,
which is extremely unlikely given that the non-test variant was _just_
called, they are pointless and provide no validation coverage.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Use KVM_IOCTL_ERROR() for one-off arm64 ioctls
Sean Christopherson [Thu, 17 Feb 2022 18:57:06 +0000 (10:57 -0800)]
KVM: selftests: Use KVM_IOCTL_ERROR() for one-off arm64 ioctls

Use the KVM_IOCTL_ERROR() macro to generate error messages for a handful
of one-off arm64 ioctls.  The calls in question are made without an
associated struct kvm_vm/kvm_vcpu as they are used to configure those
structs, i.e. can't be easily converted to e.g. vcpu_ioctl().

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Get rid of kvm_util_internal.h
Sean Christopherson [Tue, 15 Feb 2022 21:21:19 +0000 (13:21 -0800)]
KVM: selftests: Get rid of kvm_util_internal.h

Fold kvm_util_internal.h into kvm_util_base.h, i.e. make all KVM utility
stuff "public".  Hiding struct implementations from tests has been a
massive failure, as it has led to pointless and poorly named wrappers,
unnecessarily opaque code, etc...

Not to mention that the approach was a complete failure as evidenced by
the non-zero number of tests that were including kvm_util_internal.h.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Make x86-64's register dump helpers static
Sean Christopherson [Wed, 16 Feb 2022 17:14:17 +0000 (09:14 -0800)]
KVM: selftests: Make x86-64's register dump helpers static

Make regs_dump() and sregs_dump() static, they're only implemented by
x86 and only used internally.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Use __KVM_SYSCALL_ERROR() to handle non-KVM syscall errors
Sean Christopherson [Wed, 4 May 2022 16:44:49 +0000 (09:44 -0700)]
KVM: selftests: Use __KVM_SYSCALL_ERROR() to handle non-KVM syscall errors

Use __KVM_SYSCALL_ERROR() to report and pretty print non-KVM syscall and
ioctl errors, e.g. for mmap(), munmap(), uffd ioctls, etc...

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Use kvm_ioctl() helpers
Sean Christopherson [Tue, 15 Feb 2022 20:39:12 +0000 (12:39 -0800)]
KVM: selftests: Use kvm_ioctl() helpers

Use the recently introduced KVM-specific ioctl() helpers instead of open
coding calls to ioctl() just to pretty print the ioctl name.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Make kvm_ioctl() a wrapper to pretty print ioctl name
Sean Christopherson [Thu, 17 Feb 2022 18:21:36 +0000 (10:21 -0800)]
KVM: selftests: Make kvm_ioctl() a wrapper to pretty print ioctl name

Make kvm_ioctl() a macro wrapper and print the _name_ of the ioctl on
failure instead of the number.

Deliberately do not use __stringify(), as that will expand the ioctl all
the way down to its numerical sequence, again the intent is to print the
name of the macro.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: sefltests: Use vm_ioctl() and __vm_ioctl() helpers
Sean Christopherson [Tue, 15 Feb 2022 20:09:56 +0000 (12:09 -0800)]
KVM: sefltests: Use vm_ioctl() and __vm_ioctl() helpers

Use the recently introduced VM-specific ioctl() helpers instead of open
coding calls to ioctl() just to pretty print the ioctl name.  Keep a few
open coded assertions that provide additional info.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Make vm_ioctl() a wrapper to pretty print ioctl name
Sean Christopherson [Tue, 15 Feb 2022 20:06:02 +0000 (12:06 -0800)]
KVM: selftests: Make vm_ioctl() a wrapper to pretty print ioctl name

Make vm_ioctl() a macro wrapper and print the _name_ of the ioctl on
failure instead of the number.

Deliberately do not use __stringify(), as that will expand the ioctl all
the way down to its numerical sequence.  Again the intent is to print the
name of the macro.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Add vcpu_get() to retrieve and assert on vCPU existence
Sean Christopherson [Tue, 15 Feb 2022 19:31:16 +0000 (11:31 -0800)]
KVM: selftests: Add vcpu_get() to retrieve and assert on vCPU existence

Add vcpu_get() to wrap vcpu_find() and deduplicate a pile of code that
asserts the requested vCPU exists.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Remove vcpu_get_fd()
Sean Christopherson [Tue, 15 Feb 2022 19:24:01 +0000 (11:24 -0800)]
KVM: selftests: Remove vcpu_get_fd()

Drop vcpu_get_fd(), it no longer has any users, and really should not
exist as the framework has failed if tests need to manually operate on
a vCPU fd.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Use vcpu_access_device_attr() in arm64 code
Sean Christopherson [Tue, 15 Feb 2022 19:23:34 +0000 (11:23 -0800)]
KVM: selftests: Use vcpu_access_device_attr() in arm64 code

Use vcpu_access_device_attr() in arm's arch_timer test instead of
manually retrieving the vCPU's fd.  This will allow dropping vcpu_get_fd()
in a future patch.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Add __vcpu_run() helper
Sean Christopherson [Tue, 15 Feb 2022 19:49:42 +0000 (11:49 -0800)]
KVM: selftests: Add __vcpu_run() helper

Add __vcpu_run() so that tests that want to avoid asserts on KVM_RUN
failures don't need to open code the ioctl() call.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: sefltests: Use vcpu_ioctl() and __vcpu_ioctl() helpers
Sean Christopherson [Tue, 15 Feb 2022 19:05:21 +0000 (11:05 -0800)]
KVM: sefltests: Use vcpu_ioctl() and __vcpu_ioctl() helpers

Use the recently introduced vCPU-specific ioctl() helpers instead of
open coding calls to ioctl() just to pretty print the ioctl name.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Split vcpu_set_nested_state() into two helpers
Sean Christopherson [Tue, 15 Feb 2022 18:08:45 +0000 (10:08 -0800)]
KVM: selftests: Split vcpu_set_nested_state() into two helpers

Split vcpu_nested_state_set() into a wrapper that asserts, and an inner
helper that does not.  Passing a bool is all kinds of awful as it's
unintuitive for readers and requires returning an 'int' from a function
that for most users can never return anything other than "success".

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Drop @mode from common vm_create() helper
Sean Christopherson [Tue, 15 Feb 2022 01:20:17 +0000 (17:20 -0800)]
KVM: selftests: Drop @mode from common vm_create() helper

Drop @mode from vm_create() and have it use VM_MODE_DEFAULT.  Add and use
an inner helper, __vm_create(), to service the handful of tests that want
something other than VM_MODE_DEFAULT.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Make vcpu_ioctl() a wrapper to pretty print ioctl name
Sean Christopherson [Tue, 15 Feb 2022 18:15:00 +0000 (10:15 -0800)]
KVM: selftests: Make vcpu_ioctl() a wrapper to pretty print ioctl name

Make vcpu_ioctl() a macro wrapper and pretty the _name_ of the ioctl on
failure instead of the number.  Add inner macros to allow handling cases
where the name of the ioctl needs to be resolved higher up the stack, and
to allow using the formatting for non-ioctl syscalls without being
technically wrong.

Deliberately do not use __stringify(), as that will expand the ioctl all
the way down to its numerical sequence, again the intent is to print the
name of the macro.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Add another underscore to inner ioctl() helpers
Sean Christopherson [Tue, 15 Feb 2022 18:27:55 +0000 (10:27 -0800)]
KVM: selftests: Add another underscore to inner ioctl() helpers

Add a second underscore to inner ioctl() helpers to better align with
commonly accepted kernel coding style, and to allow using a single
underscore variant in the future for macro shenanigans.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Always open VM file descriptors with O_RDWR
Sean Christopherson [Tue, 15 Feb 2022 01:12:38 +0000 (17:12 -0800)]
KVM: selftests: Always open VM file descriptors with O_RDWR

Drop the @perm param from vm_create() and always open VM file descriptors
with O_RDWR.  There's no legitimate use case for other permissions, and
if a selftest wants to do oddball negative testing it can open code the
necessary bits instead of forcing a bunch of tests to provide useless
information.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Drop stale declarations from kvm_util_base.h
Sean Christopherson [Mon, 2 May 2022 22:25:35 +0000 (15:25 -0700)]
KVM: selftests: Drop stale declarations from kvm_util_base.h

Drop declarations for allocate_kvm_dirty_log() and vm_create_device(),
which no longer have implementations.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Fix typo in vgic_init test
Sean Christopherson [Wed, 16 Feb 2022 19:47:05 +0000 (11:47 -0800)]
KVM: selftests: Fix typo in vgic_init test

When iterating over vCPUs, invoke access_v3_redist_reg() on the "current"
vCPU instead of vCPU0, which is presumably what was intended by iterating
over all vCPUs.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Fix buggy-but-benign check in test_v3_new_redist_regions()
Sean Christopherson [Wed, 23 Feb 2022 22:52:12 +0000 (14:52 -0800)]
KVM: selftests: Fix buggy-but-benign check in test_v3_new_redist_regions()

Update 'ret' with the return value of _kvm_device_access() prior to
asserting that ret is non-zero.  In the current code base, the flaw is
benign as 'ret' is guaranteed to be -EBUSY from the previous run_vcpu(),
which also means that errno==EBUSY prior to _kvm_device_access(), thus
the "errno == EFAULT" part of the assert means that a false negative is
impossible (unless the kernel is being truly mean and spuriously setting
errno=EFAULT while returning success).

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: Fix references to non-existent KVM_CAP_TRIPLE_FAULT_EVENT
Sean Christopherson [Wed, 1 Jun 2022 19:16:53 +0000 (12:16 -0700)]
KVM: Fix references to non-existent KVM_CAP_TRIPLE_FAULT_EVENT

The x86-only KVM_CAP_TRIPLE_FAULT_EVENT was (appropriately) renamed to
KVM_CAP_X86_TRIPLE_FAULT_EVENT when the patches were applied, but the
docs and selftests got left behind.  Fix them.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Bug the VM on an out-of-bounds data read
Sean Christopherson [Thu, 26 May 2022 21:08:17 +0000 (21:08 +0000)]
KVM: x86: Bug the VM on an out-of-bounds data read

Bug the VM and terminate emulation if an out-of-bounds read into the
emulator's data cache occurs.  Knowingly contuining on all but guarantees
that KVM will overwrite random kernel data, which is far, far worse than
killing the VM.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20220526210817.3428868-9-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Bug the VM if the emulator generates a bogus exception vector
Sean Christopherson [Thu, 26 May 2022 21:08:16 +0000 (21:08 +0000)]
KVM: x86: Bug the VM if the emulator generates a bogus exception vector

Bug the VM if KVM's emulator attempts to inject a bogus exception vector.
The guest is likely doomed even if KVM continues on, and propagating a
bad vector to the rest of KVM runs the risk of breaking other assumptions
in KVM and thus triggering a more egregious bug.

All existing users of emulate_exception() have hardcoded vector numbers
(__load_segment_descriptor() uses a few different vectors, but they're
all hardcoded), and future users are likely to follow suit, i.e. the
change to emulate_exception() is a glorified nop.

As for the ctxt->exception.vector check in x86_emulate_insn(), the few
known times the WARN has been triggered in the past is when the field was
not set when synthesizing a fault, i.e. for all intents and purposes the
check protects against consumption of uninitialized data.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20220526210817.3428868-8-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Bug the VM if the emulator accesses a non-existent GPR
Sean Christopherson [Thu, 26 May 2022 21:08:15 +0000 (21:08 +0000)]
KVM: x86: Bug the VM if the emulator accesses a non-existent GPR

Bug the VM, i.e. kill it, if the emulator accesses a non-existent GPR,
i.e. generates an out-of-bounds GPR index.  Continuing on all but
gaurantees some form of data corruption in the guest, e.g. even if KVM
were to redirect to a dummy register, KVM would be incorrectly read zeros
and drop writes.

Note, bugging the VM doesn't completely prevent data corruption, e.g. the
current round of emulation will complete before the vCPU bails out to
userspace.  But, the very act of killing the guest can also cause data
corruption, e.g. due to lack of file writeback before termination, so
taking on additional complexity to cleanly bail out of the emulator isn't
justified, the goal is purely to stem the bleeding and alert userspace
that something has gone horribly wrong, i.e. to avoid _silent_ data
corruption.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20220526210817.3428868-7-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Reduce the number of emulator GPRs to '8' for 32-bit KVM
Sean Christopherson [Thu, 26 May 2022 21:08:14 +0000 (21:08 +0000)]
KVM: x86: Reduce the number of emulator GPRs to '8' for 32-bit KVM

Reduce the number of GPRs emulated by 32-bit KVM from 16 to 8.  KVM does
not support emulating 64-bit mode on 32-bit host kernels, and so should
never generate accesses to R8-15.

Opportunistically use NR_EMULATOR_GPRS in rsm_load_state_{32,64}() now
that it is precise and accurate for both flavors.

Wrap the definition with full #ifdef ugliness; sadly, IS_ENABLED()
doesn't guarantee a compile-time constant as far as BUILD_BUG_ON() is
concerned.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Message-Id: <20220526210817.3428868-6-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Use 16-bit fields to track dirty/valid emulator GPRs
Sean Christopherson [Thu, 26 May 2022 21:08:13 +0000 (21:08 +0000)]
KVM: x86: Use 16-bit fields to track dirty/valid emulator GPRs

Use a u16 instead of a u32 to track the dirty/valid status of GPRs in the
emulator.  Unlike struct kvm_vcpu_arch, x86_emulate_ctxt tracks only the
"true" GPRs, i.e. doesn't include RIP in its array, and so only needs to
track 16 registers.

Note, maxing out at 16 GPRs is a fundamental property of x86-64 and will
not change barring a massive architecture update.  Legacy x86 ModRM and
SIB encodings use 3 bits for GPRs, i.e. support 8 registers.  x86-64 uses
a single bit in the REX prefix for each possible reference type to double
the number of supported GPRs to 16 registers (4 bits).

Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20220526210817.3428868-5-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Omit VCPU_REGS_RIP from emulator's _regs array
Sean Christopherson [Thu, 26 May 2022 21:08:12 +0000 (21:08 +0000)]
KVM: x86: Omit VCPU_REGS_RIP from emulator's _regs array

Omit RIP from the emulator's _regs array, which is used only for GPRs,
i.e. registers that can be referenced via ModRM and/or SIB bytes.  The
emulator uses the dedicated _eip field for RIP, and manually reads from
_eip to handle RIP-relative addressing.

To avoid an even bigger, slightly more dangerous change, hardcode the
number of GPRs to 16 for the time being even though 32-bit KVM's emulator
technically should only have 8 GPRs.  Add a TODO to address that in a
future commit.

See also the comments above the read_gpr() and write_gpr() declarations,
and obviously the handling in writeback_registers().

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Message-Id: <20220526210817.3428868-4-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Harden _regs accesses to guard against buggy input
Sean Christopherson [Thu, 26 May 2022 21:08:11 +0000 (21:08 +0000)]
KVM: x86: Harden _regs accesses to guard against buggy input

WARN and truncate the incoming GPR number/index when reading/writing GPRs
in the emulator to guard against KVM bugs, e.g. to avoid out-of-bounds
accesses to ctxt->_regs[] if KVM generates a bogus index.  Truncate the
index instead of returning e.g. zero, as reg_write() returns a pointer
to the register, i.e. returning zero would result in a NULL pointer
dereference.  KVM could also force the index to any arbitrary GPR, but
that's no better or worse, just different.

Open code the restriction to 16 registers; RIP is handled via _eip and
should never be accessed through reg_read() or reg_write().  See the
comments above the declarations of reg_read() and reg_write(), and the
behavior of writeback_registers().  The horrific open coded mess will be
cleaned up in a future commit.

There are no such bugs known to exist in the emulator, but determining
that KVM is bug-free is not at all simple and requires a deep dive into
the emulator.  The code is so convoluted that GCC-12 with the recently
enable -Warray-bounds spits out a false-positive due to a GCC bug:

  arch/x86/kvm/emulate.c:254:27: warning: array subscript 32 is above array
                                 bounds of 'long unsigned int[17]' [-Warray-bounds]
    254 |         return ctxt->_regs[nr];
        |                ~~~~~~~~~~~^~~~
  In file included from arch/x86/kvm/emulate.c:23:
  arch/x86/kvm/kvm_emulate.h: In function 'reg_rmw':
  arch/x86/kvm/kvm_emulate.h:366:23: note: while referencing '_regs'
    366 |         unsigned long _regs[NR_VCPU_REGS];
        |                       ^~~~~

Link: https://lore.kernel.org/all/YofQlBrlx18J7h9Y@google.com
Link: https://bugzilla.kernel.org/show_bug.cgi?id=216026
Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105679
Reported-and-tested-by: Robert Dinse <nanook@eskimo.com>
Reported-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20220526210817.3428868-3-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Grab regs_dirty in local 'unsigned long'
Sean Christopherson [Thu, 26 May 2022 21:08:10 +0000 (21:08 +0000)]
KVM: x86: Grab regs_dirty in local 'unsigned long'

Capture ctxt->regs_dirty in a local 'unsigned long' instead of casting it
to an 'unsigned long *' for use in for_each_set_bit().  The bitops helpers
really do read the entire 'unsigned long', even though the walking of the
read value is capped at the specified size.  I.e. 64-bit KVM is reading
memory beyond ctxt->regs_dirty, which is a u32 and thus 4 bytes, whereas
an unsigned long is 8 bytes.  Functionally it's not an issue because
regs_dirty is in the middle of x86_emulate_ctxt, i.e. KVM is just reading
its own memory, but relying on that coincidence is gross and unsafe.

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20220526210817.3428868-2-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoMerge branch 'kvm-5.20-early'
Paolo Bonzini [Thu, 9 Jun 2022 15:38:12 +0000 (11:38 -0400)]
Merge branch 'kvm-5.20-early'

s390:

* add an interface to provide a hypervisor dump for secure guests

* improve selftests to show tests

x86:

* Intel IPI virtualization

* Allow getting/setting pending triple fault with KVM_GET/SET_VCPU_EVENTS

* PEBS virtualization

* Simplify PMU emulation by just using PERF_TYPE_RAW events

* More accurate event reinjection on SVM (avoid retrying instructions)

* Allow getting/setting the state of the speaker port data bit

* Rewrite gfn-pfn cache refresh

* Refuse starting the module if VM-Entry/VM-Exit controls are inconsistent

* "Notify" VM exit

2 years agoKVM: selftests: Restrict test region to 48-bit physical addresses when using nested
David Matlack [Fri, 20 May 2022 23:32:49 +0000 (23:32 +0000)]
KVM: selftests: Restrict test region to 48-bit physical addresses when using nested

The selftests nested code only supports 4-level paging at the moment.
This means it cannot map nested guest physical addresses with more than
48 bits. Allow perf_test_util nested mode to work on hosts with more
than 48 physical addresses by restricting the guest test region to
48-bits.

While here, opportunistically fix an off-by-one error when dealing with
vm_get_max_gfn(). perf_test_util.c was treating this as the maximum
number of GFNs, rather than the maximum allowed GFN. This didn't result
in any correctness issues, but it did end up shifting the test region
down slightly when using huge pages.

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220520233249.3776001-12-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Add option to run dirty_log_perf_test vCPUs in L2
David Matlack [Fri, 20 May 2022 23:32:48 +0000 (23:32 +0000)]
KVM: selftests: Add option to run dirty_log_perf_test vCPUs in L2

Add an option to dirty_log_perf_test that configures the vCPUs to run in
L2 instead of L1. This makes it possible to benchmark the dirty logging
performance of nested virtualization, which is particularly interesting
because KVM must shadow L1's EPT/NPT tables.

For now this support only works on x86_64 CPUs with VMX. Otherwise
passing -n results in the test being skipped.

Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220520233249.3776001-11-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Clean up LIBKVM files in Makefile
David Matlack [Fri, 20 May 2022 23:32:47 +0000 (23:32 +0000)]
KVM: selftests: Clean up LIBKVM files in Makefile

Break up the long lines for LIBKVM and alphabetize each architecture.
This makes reading the Makefile easier, and will make reading diffs to
LIBKVM easier.

No functional change intended.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220520233249.3776001-10-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Link selftests directly with lib object files
David Matlack [Fri, 20 May 2022 23:32:46 +0000 (23:32 +0000)]
KVM: selftests: Link selftests directly with lib object files

The linker does obey strong/weak symbols when linking static libraries,
it simply resolves an undefined symbol to the first-encountered symbol.
This means that defining __weak arch-generic functions and then defining
arch-specific strong functions to override them in libkvm will not
always work.

More specifically, if we have:

lib/generic.c:

  void __weak foo(void)
  {
          pr_info("weak\n");
  }

  void bar(void)
  {
          foo();
  }

lib/x86_64/arch.c:

  void foo(void)
  {
          pr_info("strong\n");
  }

And a selftest that calls bar(), it will print "weak". Now if you make
generic.o explicitly depend on arch.o (e.g. add function to arch.c that
is called directly from generic.c) it will print "strong". In other
words, it seems that the linker is free to throw out arch.o when linking
because generic.o does not explicitly depend on it, which causes the
linker to lose the strong symbol.

One solution is to link libkvm.a with --whole-archive so that the linker
doesn't throw away object files it thinks are unnecessary. However that
is a bit difficult to plumb since we are using the common selftests
makefile rules. An easier solution is to drop libkvm.a just link
selftests with all the .o files that were originally in libkvm.a.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220520233249.3776001-9-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Drop unnecessary rule for STATIC_LIBS
David Matlack [Fri, 20 May 2022 23:32:45 +0000 (23:32 +0000)]
KVM: selftests: Drop unnecessary rule for STATIC_LIBS

Drop the "all: $(STATIC_LIBS)" rule. The KVM selftests already depend
on $(STATIC_LIBS), so there is no reason to have an extra "all" rule.

Suggested-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220520233249.3776001-8-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Add a helper to check EPT/VPID capabilities
David Matlack [Fri, 20 May 2022 23:32:44 +0000 (23:32 +0000)]
KVM: selftests: Add a helper to check EPT/VPID capabilities

Create a small helper function to check if a given EPT/VPID capability
is supported. This will be re-used in a follow-up commit to check for 1G
page support.

No functional change intended.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220520233249.3776001-7-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Move VMX_EPT_VPID_CAP_AD_BITS to vmx.h
David Matlack [Fri, 20 May 2022 23:32:43 +0000 (23:32 +0000)]
KVM: selftests: Move VMX_EPT_VPID_CAP_AD_BITS to vmx.h

This is a VMX-related macro so move it to vmx.h. While here, open code
the mask like the rest of the VMX bitmask macros.

No functional change intended.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220520233249.3776001-6-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Refactor nested_map() to specify target level
David Matlack [Fri, 20 May 2022 23:32:42 +0000 (23:32 +0000)]
KVM: selftests: Refactor nested_map() to specify target level

Refactor nested_map() to specify that it explicityl wants 4K mappings
(the existing behavior) and push the implementation down into
__nested_map(), which can be used in subsequent commits to create huge
page mappings.

No function change intended.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220520233249.3776001-5-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Drop stale function parameter comment for nested_map()
David Matlack [Fri, 20 May 2022 23:32:41 +0000 (23:32 +0000)]
KVM: selftests: Drop stale function parameter comment for nested_map()

nested_map() does not take a parameter named eptp_memslot. Drop the
comment referring to it.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220520233249.3776001-4-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: selftests: Add option to create 2M and 1G EPT mappings
David Matlack [Fri, 20 May 2022 23:32:40 +0000 (23:32 +0000)]
KVM: selftests: Add option to create 2M and 1G EPT mappings

The current EPT mapping code in the selftests only supports mapping 4K
pages. This commit extends that support with an option to map at 2M or
1G. This will be used in a future commit to create large page mappings
to test eager page splitting.

No functional change intended.

Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220520233249.3776001-3-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>